I. Data Description
Our analysis is based on information from taxpayer returns over 11 years from the year 2001 until 2011 that the Tax Administration of the Canton of Geneva (TACG) confidentially gave us for this study. The selected variables provide information on the entire population of taxpayers in the Canton of Geneva (approximately 250’000 households). A different data set was provided for each year under study, 11 in total. Each data set comprised the same nine variables, an entire description of them is provided in Lideikyte-Huber and Pittavino (2022) and Lideikyte-Huber, Pittavino and Peter (2021); the ones particularly used in the present study are described and listed below with their original name provided in brackets. For this specific follow-up study, two new variables have specifically been created (“year” and “freqded”) to allow a more in-depth longitudinal analysis for the year of study and the characterization and computation of the frequency of donation over the study period. A merging of the 11 different data set, with the elimination of double IDs, if any, was performed to create the appropriate unique dataset:
- “coded ID” (“identifiant”): a coded ID for each taxpayers. This variable allows to follow the same taxpayer over time, except in four specific cases[1]. The same coded ID is used for a given taxpayer for each fiscal year. As Switzerland has a joint filing system, married couples are considered and treated as one taxpayer in the same way as a single non-married individual, and they have only one coded ID (in this paper, any deducting taxpayer, couple, or individual is referred to as “deducter”).
- “year of birth” (“annee_de_naissance”): the year of birth of a taxpayer (which is either an individual or a household, depending on marital status). For married couples, it is the year of birth of the “principal” taxpayer, usually the man.
- “income_bracket” (“bareme_revenu”) - the binary (0/1) indication of a possible “splitting” of income tax rate in the tax income computation, showing if a taxpayer is a couple (1) and not a single individual (0).
- “global net taxable income” (“revenu_net_imposable_taux”): the net taxable income (after all deductions) applied to set the tax rate; this includes the totality of any foreign income.
- “gross wealth” (“fortune_brute”): global gross wealth of the taxpayer.
- “deductions for donations” (“versements_benevoles”): the amount of deduction (if any) for charitable giving, representing the entire annual amount of the deducted donations (in case it is less than the deductible threshold) or capped amount of annual donations, if exceeding the deductible threshold.
- “intermediary net income for deductible donations” (“Sous_total_ded_dons”): this variable serves as a key reference point for calculating deductions that are under, equal or more than the legal threshold (10% or 20%, depending on the year). It could only be digitally extracted from the databases of the Geneva Tax Administration for the tax years 2010 and 2011. For the previous years, it was determined by internal calculations performed by the Cantonal Geneva Tax Administration (TAGC), based on the elements of the tax base that are included in its definition (Information provided by TAGC).
- “year under study” (“year”): this new variable has been generated for the purpose of this study to keep track of the evolution by year of the collected data. It indicates the 11 years under consideration for this study, from year 2001 to year 2011.
- “frequency of deductions” (“freqded”): innovative additional variable created to count the frequency a given ID repeated the charitable deductions over the study period.
This data was selected for taxpayers residing in the Canton of Geneva as well as for taxpayers residing in another Swiss canton or abroad, however still taxed in Geneva. The information above does not allow us to distinguish between these different categories of taxpayers. In addition, as from the 2009 tax year, taxpayers who are usually taxed at source (“impots à la source”, dedicated taxation practice for newcomers in Switzerland) have the possibility of filing a return, if they meet certain conditions, and are then treated as resident taxpayers (“quasi-residents”). These taxpayers are approximately 2,000 in 2009, 4,000 in 2010 and 5,600 in 2011. The variables provided by TACG does not allow us to identify quasi-resident taxpayers (Loi sur l'imposition des personnes physiques (LIPP-V)).
As reported in the previous study (Lideikyte Huber, Pittavino and Peter, 2021), the total number of taxpayers in the canton of Geneva has steadily increased, from 234,117 in 2001 to 266,336 in 2011. The share of the taxpayers deducting charitable donations more than doubled, passing from 8.3% in 2001 to 19.3% in 2011, with a steep increase in 2005 (deducting taxpayers reaching 16.3%). Concerning the general pattern of deductions during the studied period, the total amount of yearly charitable deductions increased significantly, from CHF 29,133,697 in 2001 to CHF 72,741,235 in 2011 (amounts non-adjusted for inflation) which is due to the rise in population and a substantial increase of 48% is recorded in 2009.
In the present analysis we use the terminology of “deducters” to indicate the taxpayers who contributed to charitable donations and used a tax-incentive (deduction) in relation to their donation, since we want to investigate this specific subset of taxpayers’ population.
II. Data subset and description by frequency of deducters and deductions’ ceiling
The last described variable “freqded” has been generated to allow analyzing the data with an innovative perspective by highlighting the frequency of donations from the deducters, over the 11 years under study.
Table 1 indicates the frequency of deductions by each deducter, the total number of deducters within each frequency and the resulting percentage of deducters. We observe that 29.4% of deducters (corresponding to a total of 30’319 deducters) are donating only once. The remaining 70.6% of deducters are donating more than once, showing knowledge of the tax incentives for charitable deductions, a related interest for this fiscal advantage and a start of a repeated behavior within their donations. With this targeted analysis, it was possible to identify a specific group of deducters who donated over the entire time span period of 11 years. This subgroup corresponds to 5948 taxpayers, who represents the 2.54% of the starting Geneva taxpayers’ population from 2001 - those people were continuously giving from 2001 to 2011. This subgroup of deducters will be called “deducters11” from now on and it will be compared to the subgroup of deducters identified in our previous study (Lideikyte-Huber and Pittavino, 2022), who are more interested in targeting the ceiling of deductions, referred to as ‘deducters’ subset’.
The figures related to the frequency of deductions have also been represented in the bar plot in Figure 1, where for each frequency there is shown the total number of deducters. This graph shows the decaying pattern of the frequency of deductions for all deducters, showing as the majority of people who deducted once, repeated the process in time for more than once and this reflects a repeated behavior in the deducters and an awareness of the tax incentives by the whole taxpayer’s population. However, this decaying pattern flattens considerably. It is interesting to observe that the number of donors who give very regularly, from 5 to 11 years during the studied period, is very similar. For instance, the number of donors who give once every two years and every year is nearly the same.
Table 1. Frequency of deductions, total number, and percentage of deducters.
Frequency of deductions
|
Total number
|
Percentage
|
1
|
30319
|
29.4
|
2
|
15597
|
15.1
|
3
|
10994
|
10.7
|
4
|
8530
|
8.3
|
5
|
7003
|
6.8
|
6
|
5739
|
5.6
|
7
|
5378
|
5.2
|
8
|
4677
|
4.5
|
9
|
4467
|
4.3
|
10
|
4490
|
4.4
|
11
|
5948
|
5.8
|
In Table 2 are reported the figures for the frequency of deductions for deducters interested in reaching the ceiling of 4% and more. From the third column is possible to see how 49% of deducters reaching this legal ceiling are donating only once. Moreover, among the 3.1% who are regularly donating over the 11 years under study, they represent only 5% of the totality of regular deducters “deducters11”. These first results highlight that taxpayers reaching the legal ceiling are not belonging to regular deducters. Half of taxpayers reaching the above ceiling are donating only once, as also shown in Figure 2. While from now on we will call “deducters11-ceiling” the regular deducters reaching the ceiling of donations for tax incentives.
Table 2. Frequency of deductions, total number, percentage deducters over total deducters who donated the legal ceiling of 4% and more and percentage over regular deducters.
Frequency of deductions
|
Total number
|
Percentage
|
Percentage over regular deducters
|
1
|
4667
|
49.0
|
15.4
|
2
|
1459
|
15.3
|
9.4
|
3
|
826
|
8.7
|
7.5
|
4
|
604
|
6.3
|
7.1
|
5
|
424
|
4.5
|
6.1
|
6
|
323
|
3.4
|
5.6
|
7
|
265
|
2.8
|
4.9
|
8
|
235
|
2.5
|
5.0
|
9
|
213
|
2.2
|
4.8
|
10
|
202
|
2.1
|
4.5
|
11
|
297
|
3.1
|
5.0
|
III. Deducters11 by the income bracket
The third variable described above (“splitting”) and presented in our dataset shows whether the tax payer household is entitled to specific rates that are applicable to spouses, registered partners (same-sex couples) or taxpayers who live in the same household as their minor or adult children, or a close relative who is a family dependent. By analyzing all the deducters, the deducters11 and deducters11-ceiling, this variable always shows an equally distributed population between these characteristics... even with a prevalence of joint filling. This indicates that the regularity of deductions is typical of family rather than single taxpayers, as it was instead relieved by the previous study by looking at the population’s subset features.
3. METHODS
Two different classes of statistical methods have been used with two different aims. Firstly, regression methods with and without interaction: linear bivariable models applied to two subgroups of deducters: “deducters11” and “deducters11-ceiling” to identify the main significant variables driving the charitable deductions in regular deducters, who aren’t and are reaching the ceiling threshold. Secondly, forecasting methods applied to the amount of deductions and the amount of donors, to project the predicted figures in the next upcoming 10 years: 2012-2021.
The entire data set was already analyzed in detail in Lideikyte-Huber and Pittavino (2022). For the Exploratory Data Analysis (EDA) we proceeded to focus on the new subset datasets “deducters11” and “deducters11-ceiling”. The main summary statistics (e.g., mean, SD, min, max, median) have been checked and computed. Since some of the variables represents almost the same quantity (i.e., ““global net taxable income” (“revenu_net_imposable_taux”), “intermediary net income for deductible donations” (“Sous_total_ded_dons”)), they are sharing the same part of the variance to describe the response and it resulted in a very high multicollinearity between some pairs of variables. To measure the amount of variance explained by each one of them for the resulting model the variance inflation factor (VIF) (Ref. to VIF) has been calculated. This quantity was computed to select the optimal set of variables for our analysis. This is an indication of the presence of multicollinearity. Two variables: X1: “global net taxable income” and X2: “gross wealth” resulted with an overall Mean for the VIF of 2.75.
IV. Bivariable linear regression analysis with and without interactions
The first method used to analyze the data for the frequency of deductions: “deducters11” and “deducters11-ceiling” was a bi-variable linear regression analysis between income (X1) and wealth (X2), resulted from the VIF check, with and without interaction (Faraway J. Julian, 2004 and 2016, Pittavino Marta et al, 2017a and 2017b).
Given the panel nature of our data set and the repetition over time for the frequency of deductions description bi-variable linear regression models with random effect: Linear Mixed Models (LMM) have been also performed, to incorporate better the subject variability, with and without interaction for income and wealth. LMM have been fitted to check if there was subject variability. Two types of LMM models have been implemented, with a random intercept and with random slope for each year of deductions.
We also performed robust regression analysis, with and without interaction, for both the two datasets “deducters11” and “deducters11-ceiling” to further confirm and check our findings.
v. Forecasting methods with applications in philanthropy
The second class of methods used to make predications of the current values for the next upcoming 10 years were the ETS (Error-Trend-Seasonality) and the ARIMA Models (Hyndman, 2021).
Four different type of ETS models, starting from the Simple Exponential Smoothing to Holt’s Models, that take into account different trend effects (i.e. additive, additive with damped and multiplicative) have been fitted to compute the amount of donations for the upcoming 10 years (2012-2022). Five different error metrics (i.e. AIC, AICc, BIC, RMSE and MAPE) have been calculated and the smallest one have implied to select the best model for the forecast.
For the projection of the amount of donors, an ARIMA model (Hyndman, 2021) was also fitted.